Create GNN Learner for Pathway-Aware Classification — make_gnn

Creates a Graph Neural Network that incorporates biological pathway information for omics classification.

Usage

make_gnn_learner(
  adjacency_matrix = NULL,
  n_hidden = 64L,
  n_layers = 2L,
  dropout = 0.3,
  aggregation = "mean",
  epochs = 100L
)

Arguments

adjacency_matrix: Adjacency matrix defining feature relationships (e.g., from pathway databases, correlation, or PPI networks)
n_hidden: Number of hidden units (default: 64)
n_layers: Number of GNN layers (default: 2)
dropout: Dropout rate (default: 0.3)
aggregation: Message aggregation: "mean", "sum", "max"
epochs: Number of epochs (default: 100)

Value

An mlr3 Learner object or a custom GNN wrapper

Details

The GNN uses: - Graph Convolutional layers to propagate information - Feature relationships from biological networks - Node-level predictions aggregated to graph-level

References

Kipf & Welling (2017). Semi-Supervised Classification with Graph Convolutional Networks.

Examples

if (FALSE) { # \dontrun{
if (requireNamespace("torch", quietly = TRUE) &&
    requireNamespace("mlr3torch", quietly = TRUE)) {
  # Create adjacency from correlation
  cor_mat <- cor(data)
  adj <- (abs(cor_mat) > 0.7) * 1
  diag(adj) <- 0

  learner <- make_gnn_learner(adj)
  learner$train(task)
}
} # }